10 Gigabits/second of AES256 throughput plus the CPU remains free to do it's stuff since the encryption's done on g2.2xlarge's NVIDIA GPU (in contrast, AES-NI on g2 instance gives about 2Gbps). Would this come in handy for securing large media files? What other scenarios would you find it useful?<p>A bit of quick background -- A team of 4 GPU engineers developed and then super optimized AES (all key & block sizes) on CUDA and OpenCL. On dedicated GPUs, we can go as much as 80Gbps on highend GPUs. On the ec2 g2 instances, we achieve 10Gbps. This throughput is much higher than what you would get from AES-NI (Intel's hardware accelerated AES instructions) and also save your CPU cycles for all your other stuff. It runs on CPU too if no GPU is available.<p>We built it into a cross platform (Windows/Linux) commercial SDK solution. You can plug it into your application with just a couple of lines of code change!<p>It's about 50k LOC with all GPU kernels hand optimized to squeeze every possible flop out of the GPUs. We're just limited by the PCIe data transfer limits.<p>Now I am exploring my options and thinking about selling the technology and need feedback from people who have sold source code before and pros & cons.